> If America wants to build at the speed AI requires, vertical integration isn’t optional. We’re standing up our own foundry and our own large scale CNC machining capability.
Yet China, the industrial superpower, doesn't work like that. Nothing is vertically integrated and instead a massive amount of suppliers are part of a gigantic and flexible supply-chain.
The fact that CCP's China able to have a working market of independent industrial actors, whereas Venture Capital-funded America can only works with corporation-scale central planning is an interesting paradox that I would like to have an in depth explanation for.
Hearing someone proudly describe themselves as a libertarian always reminds me of the ol’ twitter quote that goes something like “libertarians are like house cats - fiercely self-assured of their own independence, yet completely reliant on a system they don’t appreciate or understand.”
The idea that references in a scientific paper should be plentiful but aren't really that important, is a consequence of a previous technological revolution: the internet.
You'll find a lot of papers from, say, the '70s, with a grand total of maybe 10 references, all of them to crucial prior work, and if those references don't say what the author claims they should say (e.g. that the particular method that is employed is valid), then chances are that the current paper is weaker than it seems, or even invalid, and so it is extremely important to check those references.
Then the internet came along, scientists started padding their work with easily found but barely relevant references and journal editors started requiring that even "the earth is round" should be well-referenced. The result is that peer reviewers feel that asking them to check the references is akin to asking them to do a spell check. Fair enough, I agree, I usually can't be bothered to do many or any citation checks when I am asked to do peer review, but it's good to remember that this in itself is an indication of a perverted system, which we just all ignored -- at our peril -- until LLM hallucinations upset the status quo.
“There are two novels that can change a bookish fourteen-year old’s life: The Lord of the Rings and Atlas Shrugged. One is a childish fantasy that often engenders a lifelong obsession with its unbelievable heroes, leading to an emotionally stunted, socially crippled adulthood, unable to deal with the real world. The other, of course, involves orcs." -- John Rogers
You'll find a consistent set of axioms in most bigger theories. The question is if their model maps well onto the real world (which is fuzzy and inconsistent) or if it needs a lot of "if we only do a little more of X, it might finally work".
My problem with Ayn Rand is that she starts the description of her world view with agreeable statements like "A is A and therefore one can see truth (or draw objective conclusions) just by looking" (which disregards the problem of missing information). But then goes on to draw a moral from that idea which basically negates the whole point of objectivity by making the subject the center of the world. But that's not yet what makes her work propaganda.
What makes her work propaganda is that she, from there "induced" that, since there's only the individual that matters, it is only moral that one tries to maximize one's own happiness and that a fully capitalist society without any regulations whatsoever were we worship the then-to-be-godlike individual entrepreneur would be the only way to achieve said happiness. This also implies that while is is only moral to strive for one's own happiness, there is only a certain kind of individual who actually deserves it. The rest is there to worship or just be screwed over and over again, because there can be only a few winners.
So we've come a long way from making seemingly agreeable statements to justifying a system that dehumanizes most of its subjects (pun not intended) and makes them nothing more but a fleshy mass to the disposal of a select few winners. And that's just what propaganda does: drawing conclusions from a seemingly agreeable standpoint in a way that seems to be logical, but in its essence ignores the fuzziness and incompleteness of the world for the sake of some sense of purity. Don't be fooled by that. There's always complexity hiding somewhere. And while A might seem to be A, you just don't know, how large the hidden b is, yet.
In practice, I'd recommend to look for mental tools that help you analyze but always leave room to deal with the inconsistency of reality. Outside of formal science, consistency is a trap. Building a world view from a set of basic axioms works for mathematics, but not for the extremely complex network that is human relations. I had to learn that the hard way. I'd recommend thinking in networks, path dependencies, path probabilities and network centrality (power) instead. It leads you down a path that allows you to form a much clearer critique than you ever could by adopting Ayn Rand's way of thinking.
> I was only talking about the US. I am not familiar with European problems,
You're not familiar with US realities either, but nonetheless my original comment was about the whole West (which has faced the same neoliberal switch, in roughly the same time frame, with roughly the same social consequences even though the details differ a bit).
> Then, Detroit in fact has a massive number of abandoned houses, so yes prices did effectively go to zero. Presumably if they could be sold to anyone they would be, but they are not worth owning https://detroitmi.gov/news/deputy-mayor-detroit-land-bank-au...
> Thanks for providing me with this great argument for supply and demand, I will use it in future ;)
This is a fascinating way of misconstruing the reality to fit in your ideological views.
People did not stop paying rents in Detroit during the period, the effective cost of housing never reached zero, in fact people were still spending 30% of their income in housing[1] despite the houses themselves being practically worthless. That's a great example of market failure, because the market price kept hovering way above the actual market value of the goods (which is what they ended up abandoned in the first place, they were never on sale for $100 each or even $10k, they were just kept empty until they got destroyed due to lack of maintenance).
Yet you still find a way to twist that evidence into a confirmation of your beliefs.
This is genuinely fascinating. The only comparable thing I've seen is this: https://www.youtube.com/watch?v=SrGgxAK9Z5A (though in fairness the guy seems to have quite a bit more self-awareness than you do).
> Oddly, it's the thinking advocated by many HN posts, denigrating the innovation under discussion as impossible, useless, etc.
A significant fraction of HN has been raised with the idea that “natural” innovation can only arise from the private sector competing on a market, and every attempt at public-funded out-of-market innovation is seen as “unnatural” and doomed to fail.
And like all religion, it's pretty hopeless to refute it with rational arguments.
Those high-powered toys by the cops are merely for showing off and to victimize the weak.
Those toys typically never come into play to protect the citizens.
Case in point: during the Uvalde school shooting incident in 2022, when a shooter (Salvador Ramos) went on a killing spree inside the school, then hundreds of cops gathered outside
with brand new body armor (gifted to them just months ago) and armed with automatic guns, but they never dared to go inside to tackle the shooter. Not only that, those cowardly cops actively prevented parents and state patrol officers from going in to rescue their kids. The cowardly cops were led by a cowardly police chief, who later gave excuses for the delayed response to the deadly situation and his mishandling of the police force, by claiming to have forgotten his walkie talkie!
Ultimately one of the border patrol officers and some US deputy marshalls (who had travelled 70 miles to reach the scene after getting an alert) managed to sneak in to the back, break the locked door, and used a tactical shield to corner and finally kill the shooter, thus ending his bloodbath (19 children and 2 teachers were tragically killed).
And if you think arming cowardly showoff cops with guns and armor is useless and potentially dangerous, you should know the Uvalde school shooter was a minor but he managed to buy the guns legally from a gun shop on credit!
That's how lax and evil the gun laws and resulting shootouts in USA are.
USA has more mass shootings and more school shootings than any other place in the world.
No wonder they facilitate and glorify high-speed car chases. It is all a thrillride for these adrenaline junkies high on power.
I can see two problems causing the pain described here, which I will discuss shortly. But the article seems to stretch that experience too much into the 'regulation is bad' territory. Regulations exist for a reason. They aren't created for the power trip of government officials. This is the same US where companies dump PFAS into drinking water sources with impunity, has some of the highest fees for the worst quality interest access, where insulin is unaffordable and corporate house renting is a thing. There are many such areas where regulation and oversight is woefully inadequate, much less any 'overregulation'. Regulations are practically the only thing standing between the rich and the powerful and their incessant attempt to drive even more wealth into their own pockets at the expense ordinary people's health, wealth, future, welfare, housing, etc.
Now let's look at the specific problems here with a much narrower scope than 'regulations'. The first problem is the type of regulations. Some regulations are too arcane and don't reflect the current state of technology. Others affect the unprivileged people disproportionately. The solution for that is to amend these regulations fast enough - not deregulation. It's also important to assess the negative impacts of loosening these regulations - something I don't see discussed in this article.
The other important requirement is to increase the staffing of the regulatory agencies so that their individual workload doesn't become a bottleneck in the entire process. There is a scientific method to assess the staffing requirements of public service institutions. According to that, a significant number of government departments all over the world are understaffed. Regulatory agencies and police departments top that list. Increased workload on their officials lead to poor experience for the citizens availing their services (this is very evident in policing). Yet those same experiences are misconstrued and misrepresented to call for deregulation and defunding of these institutions - the opposite of what's actually needed. (PDs need more staff and more training in empathy. Not defunding, nor militarization.) This is exactly what I see in this article. An attempt to target regulations as a whole using a sob anecdote.
That sounds like what Varoufakis planned to implement in Greece in case of a default in 2015.
” What we planned to do was this. There's the tax office website, as in Britain and everywhere else, where citizens (taxpayers accessing the website) use their tax identification number and transfer money from their bank account to their tax identification number via online banking to pay VAT, income tax, and so on. We were planning to surreptitiously create reserve accounts linked to each tax identification number without notifying anyone, simply so that this system would operate in secret. With the push of a button, we could assign PIN numbers to the holders of the tax identification numbers (taxpayers). For example, in a case where the state owed a pharmaceutical company one million euros for medicines purchased on behalf of the National Health Service, we could immediately make a transfer to the reserve account corresponding to the pharmaceutical company's tax identification number and provide them with a PIN number. They could use it as a kind of parallel payment mechanism to transfer any portion of those digital funds they wanted to any tax identification number they owed money to. Or even to use it to make tax payments.”
You're thinking of 'back in the day.' The original XBox's video card was worth more than they sold the entire system for, and the PS3 was a complete beast of computation (even if not entirely inappropriate for games...)! But in modern times (PS4 gen onward) consoles have become relatively vanilla midrange computers designed with the intent of turning profit on the hardware as quickly as possible.
The hardware cost of the PS4 was less than it's retail price from day 0 [1], and they began making a profit per unit shortly thereafter. Similarly the PS5 also reached profit per unit in less than a year. [2] XBox models from the PS4 gen onward are conspicuously similar as well.
The moment we learned ChatGPT helped a teen figure out not just how to take their own life but how to make sure no one can stop them mid-act, we should've been mortified and had a discussion.
But we also decided via Sandy Hook that children can be slaughtered on the altar of the second amendment without any introspection, so I mean...were we ever seriously going to have that discussion?
This is tangential to the main point of the article, but this concluding sentence annoyed me greatly:
> Governments are generally better at supporting companies in established markets where innovation takes place slowly and incrementally. This is likely why state-backed efforts have found it easier to be competitive against aerospace companies than Silicon Valley giants working at breakneck pace
I always finds it fascinating that companies like Facebook, Google, Amazon or Microsoft (granted the later isn't a “silicon valley giant” proper) managed to build a narrative portraying themselves as “innovative companies” when they are the opposite of that: they are, and have been for almost two decades now or even more for Microsoft, very close to the complacent and short-term-profit-maximizer Boeing portrayed in this article.
They just happen to benefit from a much stronger network effect than Boeing, and work in a business where economies of scale are insane.
In safety-critical systems, we distinguish between accidents (actual loss, e.g. lives, equipment, etc.) and hazardous states. The equation is
hazardous state + environmental conditions = accident
Since we can only control the system, and not its environment, we focus on preventing hazardous states, rather than accidents. If we can keep the system out of all hazardous states, we also avoid accidents. (Trying to prevent accidents while not paying attention to hazardous states amounts to relying on the environment always being on our side, and is bound to fail eventually.)
One such hazardous state we have defined in aviation is "less than N minutes of fuel remaining when landing". If an aircraft lands with less than N minutes of fuel on board, it would only have taken bad environmental conditions to make it crash, rather than land. Thus we design commercial aviation so that planes always have N minutes of fuel remaining when landing. If they don't, that's a big deal: they've entered a hazardous state, and we never want to see that. (I don't remember if N is 30 or 45 or 60 but somewhere in that region.)
For another example, one of my children loves playing around cliffs and rocks. Initially he was very keen on promising me that he wouldn't fall down. I explained the difference between accidents and hazardous states to him in childrens' terms, and he realised slowly that he cannot control whether or not he has an accident, so it's a bad idea to promise me that he won't have an accident. What he can control is whether or not bad environmental conditions lead to an accident, and he does that by keeping out of hazardous states. In this case, the hazardous state would be standing less than a child-height within a ledge when there is nobody below ready to catch. He can promise me to avoid that, and that satisfies me a lot more than a promise to not fall.
Anyone who has lived through a market correction (the tariff announcements in early April this year being a recent example, though there have been far worse) should be able to see that market prices do not always accurately reflect even the consensus view of value (which itself can be wrong). As people are forced to de-lever, everything goes down at once, often by very similar amounts, even though it cannot be possible that everything suddenly lost the same amount of value simultaneously.
To quote Richard Bookstaber, "The principal reason for intraday price movement is the demand for liquidity... the role of the market is to provide immediacy for liquidity demanders. ...market crises... are the times when liquidity and immediacy matter most. ...the defining characteristic is that time is more important than price. ...diversification strategies fail. Assets that are uncorrelated suddenly become highly correlated, and all positions go down together. The reason for the lack of diversification is that in a high-energy market, all assets in fact are the same.... What matters is who holds the assets." (from A Framework for Understanding Market Crises, 1999)
Was the market drop an accurate reflection of the value that would have been destroyed by those tariffs, discounted by the probability that they would have been enacted as drafted? Nobody knew then, and I maintain that nobody even knows now. That was not the calculation that was being made.
A lot of banks/credit unions offer 0% loans to federal employees during government shutdowns. Some of them even offer this to customers who join after the shutdown began.
Granted, it is weird to think about living in a country where this is a finanical product that so many institutions just have.
Even without that, the shutdown actually needs to last a while before it actually becomes a problem for forloughed workers.
Despite the shutdown, the October 1 paychecks (covering work done between September 7 and September 20) will still go out, as they have long since been sent to the payment processors.
The October 15 paychecks (covering September 21 - October 4) should go out with a slightly reduced amount the unpaid 4 days during the shutdown. This would require the agency in question to properly follow shutdown protocols and submit timecards/payroll data to the processors prior to the shutdown. Since this is not the routine process, I imagine some mistakes will be made here, so some government workers will probably miss the October 15 paycheck. Others may face hardship from the reduced payment; but even people living paycheck to paycheck typically stretch the paycheck out across the pay period, so will be able to get through most of the period on the reduced check (and can probably defer/reduce some spending).
Major payroll lapses will not kick in until the October 29 paycheck. If we reach that point, this would be the second longest shutdown in history [0]. Even then, most Americans have an existing line of credit that offers 0% interest loans for between 30 and 60 days depending on when they are in the billing cycle [1].
[0] Not that I would be too suprised by this. The longest shutdown occured under President Trump, while the republicans controlled the Senate; and had controll of the House for the first part of the shutdown.
[1] Except for those who have a revolving balance, in which case and additional spending starts accruing interest immedietly; and any delay in paying down the existing balance also incurs interest. And, when these loans do gather interest, it is at a quite high rate.
>that all model makers actually have been training models explicitly on the most common tasks people use them for via synthetic data generation
People really don't understand that part in general.
I find the easiest way to make people understand is to write gibberish that will trigger the benchmaxxed "pattern matching" behavior like this:
> The child and the wolf try to enjoy a picnic by the river but there's a sheep. The boat needs to connect nine dots over the river without leaving the water but gets into an accident and dies. The surgeon says "I can't operate on this child!" why?
The mix and matching of multiple common riddles/puzzles style questions into a singular gibberish sentence should, if models had legitimate forms of reasoning, make the model state that this is nonsense, at best, or answer chaotically at worst. Instead, they will all answer "The surgeon is the mother" even though nobody even mentioned anything about anyone's gender. That's because that answer, "the surgeon is the mother", for the gender bias riddle has been burned so hard into the models they cannot reply in any other way as soon as they pattern match "The surgeon can't operate on this child". No matter how much crap you wrote before that sentence. You can change anything about what comes before "The surgeon" and the model will almost invariably fall into giving an answer like this one (Gemini 2.5 pro):
>The details about the wolf, the sheep, the picnic, and the dying boat are all distractions (red herrings) to throw you off. The core of the puzzle is the last sentence.
>The surgeon says, "I can't operate on this child!" because the surgeon is the child's mother.
One could really question the value, by the way, of burning the answer to so many useless riddles into LLMs. The only purpose it could serve is gaslighting the average person asking these questions into believing there's some form of intelligence in there. Obviously they fail so hard to generalize on this (never working quite right when you change an element of a riddle into something new) that from a practical use point of view, you might as well not bother have this in the training data, nobody's going to be more productive because LLMs can act as a database for the common riddles.
You are talking about spatial safety. There are a few other types of memory safety:
- temporal safety (e.g. no use after free)
- initialization safety (no read of initialized memory)
- thread safety (no data races)
- type safety (accessing memory with the correct type)
Every line of code is a liability, in the accounting sense of the word. That is, it's what makes the asset (software) exist.
The balance sheet is always equal, so you always have as much liability as asset in amount. Despite the negative connotation of the word, some liabilities are good (earnings[1] is the best, equity and long term debt with low interest rate are nice too), but some are bad.
And it's the same for code, spitting more code to do the same thing is akin to emitting too many shares, having brittle code that needs constant intervention to keep working is akin to high interest debt.
[1] yes, your company's earning belong to the right side of its balance sheet.
I have an overwhelming feeling that what we're trying to do here is "Netflix over DialUp."
We're clearly seeing what AI will eventually be able to do, just like many VOD, smartphone and grocery delivery companies of the 90s did with the internet. The groundwork has been laid, and it's not too hard to see the shape of things to come.
This tech, however, is still far too immature for a lot of use cases. There's enough of it available that things feel like they ought to work, but we aren't quite there yet. It's not quite useless, there's a lot you can do with AI already, but a lot of use cases that are obvious not only in retrospect will only be possible once it matures.
Very detailed and accurate description. The author clearly knows way more than I do, but I would venture a few notes:
1. In the cloud, it can be difficult to know the NUMA characteristics of your VMs. AWS, Google, etc., do not publish it. I found the ‘lscpu’ command helpful.
2. Tools like https://github.com/SoilRos/cpu-latency plot the core-to-core latency on a 2d grid. There are many example visualisations on that page; maybe you can find the chip you are using.
3. If you get to pick VM sizes, pick ones the same size as a NUMA node on the underlying hardware. Eg prefer 64-core m8g.16xlarge over 96-core m8g.24xlarge which will span two nodes.
Your threshold theory is basically Amara's Law with better psychological scaffolding. Roy Amara nailed the what ("we tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run") [1] but you're articulating the why better than most academic treatments. The invisible-to-researchers phase followed by the sudden usefulness cascade is exactly how these transitions feel from the inside.
This reminds me of the CPU wars circa 2003-2005. Intel spent years squeezing marginal gains out of Pentium 4's NetBurst architecture, each increment more desperate than the last. From 2003 to 2005, Intel shifted development away from NetBurst to focus on the cooler-running Pentium M microarchitecture [2]. The whole industry was convinced we'd hit a fundamental wall. Then boom, Intel released dual-core processors under the Pentium D brand in May 2005 [2] and suddenly we're living in a different computational universe.
But teh multi-core transition wasn't sudden at all. IBM shipped the POWER4 in 2001, the first non-embedded microprocessor with two cores on a single die [3]. Sun had been preaching parallelism since the 90s. It was only "sudden" to those of us who weren't paying attention to the right signals.
Which brings us to the $7 trillion question: where exactly are we on the transformer S-curve? Are we approaching what Richard Foster calls the "performance plateau" in "Innovation: The Attacker's Advantage" [4], where each new model delivers diminishing returns? Or are we still in that deceptive middle phase where progress feels linear but is actually exponential?
The pattern-matching pessimist in me sees all the classic late-stage S-curve symptoms. The shift from breakthrough capabilities to benchmark gaming. The pivot from "holy shit it can write poetry" to "GPT-4.5-turbo-ultra is 3% better on MMLU." The telltale sign of technological maturity: when the marketing department works harder than the R&D team.
But the timeline compression with AI is unprecedented. What took CPUs 30 years to cycle through, transformers have done in 5. Maybe software cycles are inherently faster than hardware. Or maybe we've just gotten better at S-curve jumping (OpenAI and Anthropic aren't waiting for the current curve to flatten before exploring the next paradigm).
As for whether capital can override S-curve dynamics... Christ, one can dream.. IBM torched approximately $5 billion on Watson Health acquisitions alone (Truven, Phytel, Explorys, Merge) [5]. Google poured resources into Google+ before shutting it down in April 2019 due to low usage and security issues [6]. The sailing ship effect (coined by W.H. Ward in 1967, where new technology accelerates innovation in incumbent technology)[7] si real, but you can't venture-capital your way past physics.
I think we can predict all this capital pouring in to AI might actually accelerate S-curve maturation rather than extend it. All that GPU capacity, all those researchers, all that parallel experimentation? We're speedrunning the entire innovation cycle, which means we might hit the plateau faster too.
You're spot on about the perception divide imo. The overhyped folks are still living in 2022's "holy shit ChatGPT" moment, while the skeptics have fast-forwarded to 2025's "is that all there is?" Both groups are right, just operating on different timescales. It's Schrödinger's S-curve where we things feel simultaneously revolutionary and disappointing, depending on which part of the elephant you're touching.
The real question I have is whether we're approaching the limits of the current S-curve (we probably are), but whether there's another curve waiting in the wings. I'm not a researcher in this space nor do I follow the AI research beat to weigh in but hopefully someone in the thread can? With CPUs, we knew dual-core was coming because the single-core wall was obvious. With transformers, the next paradigm is anyone's guess. And that uncertainty, more than any technical limitation, might be what makes this moment feel so damn weird.
> The internet evolved from the telegraphy network, which was not created by the government.
Claining that the internet evolved from telegraph is very much bad faith, but the worse part about this argument is that it's wrong: the first country-scale telegraph network was indeed funded by the French government[1].
> As for bad environmental effects, socialism has a much worse track record
Why are you guys obsessed so much with socialism? There is no mention of socialism anywhere in this thread. Government funding stuff has nothing to do with socialism in the first place, otherwise it would mean that every developed country is a socialist one, with the only non-socialist countries being failed states like Somalia which really isn't the argument you want to make.
> Free markets produce enough surplus that costly mitigations become practical.
“Free market” doesn't exist, it's a propaganda phrase with no basis on reality. The government always and everywhere has a key role to play in the economy, by counteracting all kinds of negative outcomes that arise from markets (mitigating crashes or maintaining consumer trust through regulations to name a few).
If it was, then the housing price would be correlated with the level of regulation, but it's not (housing less dense places is always cheaper than housing in big cities, even if you compare the most heavily regulated place on earth with the least regulated big city in the world.
> Here's how a rational market works.
A rational market made from human is as realistic as “dry water”.
The key issue is that housing is treated as an asset, and the policies made to boost housing construction are designed around building a market in a way that gives housing the same kind of financial yields as stocks. But while stock prices can grow because productivity increase, housing value can only increase if housing becomes more expensive.
Time[1] and time[2] again, the CJEU has ruled that the US stance of noncitizens having no standing on privacy issues is incompatible with EU law. Time[3] and time[4] again, the European Commission has negotiated a functionally identical agreement codified in executive orders and declared it “adequate” until the court could decide otherwise. Not even the Congress explicitly giving[5] the US government powers to compel (among others) EU subsidiaries of US multinationals, regardless of what EU law says, has changed the equation. Now there’s been a presidential election in the US that many in the EU are unhappy about. *Shocked Pikachu*
> [French MP Philippe] Latombe criticised the US-EU Data Privacy Framework (DPF) deal, saying it no longer served EU interests due to the US president’s “impulsive” nature.
Am I wrong to say that there’s something profoundly rotten in that statement with regards to the rule of law?
> If America wants to build at the speed AI requires, vertical integration isn’t optional. We’re standing up our own foundry and our own large scale CNC machining capability.
Yet China, the industrial superpower, doesn't work like that. Nothing is vertically integrated and instead a massive amount of suppliers are part of a gigantic and flexible supply-chain.
The fact that CCP's China able to have a working market of independent industrial actors, whereas Venture Capital-funded America can only works with corporation-scale central planning is an interesting paradox that I would like to have an in depth explanation for.